A building collapse assessment method based on stereoscopic imagery and laser altimetry data

Abstract

The application discloses a building collapse evaluation method based on stereoscopic images and laser height measurement data, adopts a coarse-to-fine matching method to match a plurality of stereoscopic images before and after an earthquake and construct a connection point error equation, simultaneously matches internal laser height measurement data and stereoscopic images to realize extraction of laser points and construction of an internal laser point error equation, matches external laser height measurement data and stereoscopic images to construct an external laser point error equation, and finally forms a adjustment model; the adjustment model is solved, regional network adjustment is completed, and a compensation coefficient corresponding to each stereoscopic image is obtained; a post-adjustment rational function imaging model corresponding to each stereoscopic image is obtained; corner points of a building are automatically measured, image coordinates corresponding to each corner point are obtained, the image coordinates are input into the post-adjustment rational function imaging model of the stereoscopic image, height changes of the building are calculated, and evaluation of a collapse degree of the building is realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of image processing, and particularly relates to a building collapse evaluation method based on stereoscopic images and laser height measurement data. BACKGROUND

[0002] Traditional disaster evaluation can obtain two-dimensional information such as the position and area of damaged houses, but it is difficult to evaluate the change of three-dimensional information of the houses. The stereoscopic satellite has the characteristics of wide range and short response period, and can calculate the change of three-dimensional information of the houses, which is very suitable for the fast response and wide range required by the earthquake disaster evaluation work. Using HRSI to perform earthquake disaster evaluation needs to improve the vertical and plane positioning accuracy of HRSI through ground control points. The specific method is to register the ground control points with the HRSI and construct a control point error equation, extract the connection points between the HRSI images and construct a connection point error equation, and solve the error equation to obtain the compensation coefficient of the stereoscopic image. The positioning accuracy of the satellite stereoscopic image seriously depends on the ground resolution of the stereoscopic image itself and the layout of the ground control points. However, due to the large difference between the pre-earthquake and post-earthquake environment, it is very difficult to quickly obtain the ground control points, which leads to the inconsistency of the plane and vertical positioning accuracy between the pre-earthquake and post-earthquake stereoscopic images, and the problems such as difficulty in obtaining the damaged house corner points during the house damage evaluation, thereby leading to the inaccuracy of the house evaluation result. SUMMARY

[0003] In view of the problems such as inconsistent plane positioning, insufficient height positioning accuracy, difficulty in laying ground control points after the disaster, and difficulty in selecting the image side of the damaged house, which are encountered in the evaluation of the damaged house using HRSI, the application provides a building collapse three-dimensional refined evaluation method based on pre-earthquake and post-earthquake stereoscopic images of GF-7 and laser height measurement data. Firstly, the pre-earthquake and post-earthquake stereoscopic images are inconsistent in plane positioning and insufficient in vertical positioning accuracy, and GF-7 and ICESat1 laser height measurement data are used as height control and fourfold connection points before and after the earthquake are used as constraints for regional network adjustment. Secondly, in view of the radiation difference between the pre-earthquake and post-earthquake stereoscopic images and the great difference between the damaged house corner points before and after the earthquake, a coarse-to-fine sub-pixel house corner point extraction method is proposed. The method first performs coarse registration by using homography matrix and epipolar constraint, and then performs fine registration by using least squares template matching to obtain the image side of the house corner points in the pre-earthquake and post-earthquake stereoscopic images. Finally, the height change of the house corner points is calculated according to the pre-earthquake and post-earthquake stereoscopic images, and the three categories of EMS are divided to perform building collapse three-dimensional refined evaluation.

[0004] The application can be implemented by the following technical solutions:

[0005] A building collapse evaluation method based on stereoscopic images and laser height measurement data, comprising the following steps:

[0006] Step one, adopt the method of coarse to fine matching to match multiple stereo images before and after the earthquake, and combine the affine model to construct the connection point error equation, at the same time, match the internal laser height data, namely the footprint image and the stereo image to realize the extraction of laser points and combine the affine model to construct the internal laser point error equation, and match the external laser height data and the stereo image to construct the external laser point error equation, and jointly construct the adjustment model based on the rational function imaging model;

[0007] Step two, solve the adjustment model, complete the regional network adjustment and obtain the compensation coefficient corresponding to each stereo image, so as to obtain the adjusted rational function imaging model corresponding to each stereo image;

[0008] Step three, adopt the coarse to fine positioning method to realize the automatic measurement of building corner points and obtain the image coordinates corresponding to each corner point;

[0009] Step four, input the image coordinates in step three into the adjusted rational function imaging model of the stereo images before and after the earthquake respectively, calculate the height change of the building, and realize the evaluation of the collapse degree of the building.

[0010] Further, the method of constructing the connection point error equation in step one includes the following steps:

[0011] Step I, select one of the multiple stereo images as the main image, and the others as the auxiliary images, and use the SIFT algorithm to extract the feature points of the main image;

[0012] Step II, calculate the longitude and latitude corresponding to each feature point on the main image, take the target feature point as the target feature point, input the rational function imaging model of the other auxiliary images combined with the maximum and minimum elevation values of the main image, obtain the image block containing the target feature point, and then take the target feature point on the main image as a template, adopt the least square matching algorithm to calculate the fine coordinates of the target feature point on the other auxiliary images;

[0013] Step III, take the target feature point of the main image and the matching points corresponding to the other auxiliary images as the connection points, and construct the connection point error equation.

[0014] Further, when using the SIFT algorithm to extract the feature points of the main image in step I, first divide the main image into blocks, then use SIFT to extract the feature points of each image block, and then use the ANMS algorithm to uniformize the feature points in each image block, and retain the feature points with a response value greater than 20%.

[0015] Further, the method of constructing the internal laser point error equation in step one includes the following steps:

[0016] Step I, using SIFT to extract feature points from the footprint image;

[0017] Step II, calculating the longitude and latitude of each feature point on the footprint image, taking the target feature point as the target, combining the maximum and minimum elevation values of the footprint image, inputting the rational function imaging model of each stereo image to obtain the image block containing the target feature point, and then taking the target feature point of the footprint image as a template to calculate the fine coordinates of the target feature point on each stereo image by using the least square matching algorithm;

[0018] Step III, calculating the homography matrix from the footprint image to each stereo image from the matching point pairs between the footprint image and each stereo image, inputting the image coordinates of the laser points on the footprint image into the matrix to obtain the image coordinates of the laser points on the stereo images, and constructing the internal laser point error equation.

[0019] Further, the external laser height measurement data is projected onto each stereo image, and then the coarse registration of the laser height measurement data is performed by using the corresponding homography matrix constructed according to the connecting points obtained in the foregoing, and then the fine registration of the laser height measurement data is performed by using the least square template matching on all stereo images, respectively, to construct the external laser point error equation by using the matching points corresponding to the external laser height measurement data and each stereo image.

[0020] Further, the adjustment model based on the rational function imaging model in the step I is constructed as follows:

[0021]

[0022] wherein the first type of equation is the connecting point error equation, the second type is the virtual control point error equation, and the third type is the laser point error equation composed of the internal and external, V tp , V vcp , V las respectively represent the error terms of the connecting points, the virtual control points and the laser points, A tp , A vcp , A las respectively represent the partial derivative terms of the connecting points, the virtual control points and the laser points with respect to the undetermined solving coefficients in the affine transformation compensation model, X aff is the compensation coefficient to be solved, B tp , B vcp , B las respectively represent the partial derivative terms of the connecting points, the virtual control points and the laser points with respect to the object points, X tp , X vcp , X las respectively represent the object terms of the connecting points, the virtual control points and the laser points, L tp , L vcp , L lasResidual terms representing the connection points, virtual control points and laser points, respectively, P tp Residual terms representing the connection points, virtual control points and laser points, respectively, P vcp Residual terms representing the connection points, virtual control points and laser points, respectively, P las Weight terms representing the connection points, virtual control points and laser points, respectively.

[0023] Further, the automatic measurement of the building corner points in step three is as follows: firstly, the fine image coordinates of the building corner points are selected by visual interpretation on the main image, a corresponding homography matrix is constructed in combination with the connection points obtained according to the above description, and the coarse positioning of the building corner points on other sub-images is realized; then, the fine image coordinates of the building corner points on the main image are taken as a template, a least square matching algorithm is adopted, and the points with a correlation coefficient greater than 0.8 are reserved to automatically calculate the fine image coordinates of the building corner points on other sub-images.

[0024] The present application has the beneficial technical effects in that:

[0025] The laser height measurement data is used, and the plane positioning accuracy of the pre-seismic and post-seismic stereo images is unified, the time-consuming and labor-consuming ground control points which are not conducive to the timeliness requirement of the earthquake disaster assessment work are avoided, only the vertical positioning accuracy is considered, and the automatic registration of the laser height measurement data is completed. In the controlled case, compared with the positioning of the pre-seismic stereo image alone, after unifying the pre-seismic and post-seismic positioning accuracy, the height accuracy is obviously improved by 52%, the vertical accuracy RMSE is improved from 2m to 0.98m, and the image internal accuracy RMSE is improved from 0.63 to 0.33; at the same time, in the uncontrolled case, the positioning accuracy of the pre-seismic image is improved by 82% after unifying the pre-seismic and post-seismic positioning accuracy, the vertical accuracy RMSE is improved from 10.28m to 2.3m, and the image internal accuracy RMSE is changed from 0.57 to 0.65.

[0026] For building corner point measurement, the automatic measurement method from coarse to fine based on visual interpretation, homography matrix and least square template matching can complete the automatic measurement of multiple building corner points before and after the earthquake, has good measurement effect, greatly saves manpower, and avoids the problem of uneven measurement accuracy of different individuals. Four seriously damaged areas in the disaster area are selected for evaluation, the accuracy reaches 93.6% in area A, 47% in area B, 97.3% in area C, and 100% in area D, which meets the accuracy requirement of the evaluation of the damage of houses in the disaster area. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 It is a schematic diagram of the overall process of the present application;

[0028] Figure 2 It is a schematic diagram of the same point extraction method of the present application for overcoming the difference in different time phases;

[0029] Figure 3A flowchart of a matching process of the footprint image and the stereoscopic image of the present application;

[0030] Figure 4 A flowchart of the automatic acquisition of the house corner point of the present application.

[0031] Figure 5 A schematic diagram of the evaluation result for a single piece region of the present application.

[0032] Figure 6 A schematic diagram of the A, B, C, D four changed house regions selected by visual interpretation of the present application. DETAILED DESCRIPTION

[0033] The specific embodiments of the present application will be described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0034] As Figure 1 shown, the present application provides a building collapse evaluation method based on stereoscopic images and laser height measurement data. Based on SIFT and least square matching, the radiation difference of multiple stereoscopic images before and after the earthquake can be overcome to extract the same name points. After combining with external laser height measurement auxiliary data, the geometric positioning inconsistency before and after the earthquake is eliminated through block adjustment, the three-dimensional space vertical positioning accuracy is improved, and the relative plane positioning error is eliminated. Then, the forward intersection of the stereoscopic images before and after the earthquake is realized to evaluate the height change of the house corner point. Then, according to the building standard of the disaster area, the damage degree of each building in the disaster area is evaluated, which is as follows:

[0035] Step one, a coarse-to-fine matching method is used to match multiple stereoscopic images before and after the earthquake and construct a connection point error equation. For each of the two stereoscopic images before and after the earthquake, the internal laser height measurement data, i.e. the footprint image, is matched with the stereoscopic image to realize the extraction of laser points and construct an internal laser point error equation. In addition, the external laser height measurement data is matched with the stereoscopic image to construct an external laser point error equation.

[0036] The stereoscopic image data before and after the earthquake is different time phase data collected at different times, which has the problem of inconsistent reference. Under different time phases, the satellite orbit positioning, attitude, and subsatellite point positioning parameters are different, which leads to great differences in the imaging range, horizontal, and height positioning accuracy of different time phase stereoscopic images. Based on the number of projection rays before and after the earthquake, different error equations are constructed for the connection points, laser points, and virtual control points, forming more strict constraints to obtain more accurate positioning.

[0037] 1. Same name point extraction of stereoscopic images before and after the earthquake to construct connection points and connection point error equations

[0038] The related research shows that the distribution of connection points has a direct impact on the elimination of the relative positioning error between stereo images, therefore, we take multiple stereo images before and after the earthquake as research data, select one as the main image and the rest as the auxiliary images, as shown in Figure 2 Firstly, the main image is divided into blocks, and the SIFT algorithm is used to extract feature points from each image block, so that the feature points distributed on the whole image can be obtained, and then the ANMS algorithm is used to uniformize the feature points in a single image block, and the feature points with a response value greater than 20% are reserved.

[0039] Then, a coarse-to-fine strategy is used to obtain the homonymic points of the existing feature points in other auxiliary images, that is, for each feature point of the main image, the longitude and latitude of the target feature point are roughly calculated by using the ground resolution and the row and column numbers of the main image, and the maximum and minimum elevation values (which can be calculated by RPCs) of the main image are combined with the rational function imaging model of other auxiliary images, so that the image block containing the target feature point can be calculated on each auxiliary image, and the coarse positioning is completed.

[0040] Finally, the least squares matching algorithm is used to calculate the fine coordinates of the target feature points in other auxiliary images by taking the feature points on the main image as a template, so as to complete the extraction of the homonymic points.

[0041] The connection points are composed of each feature point of the main image and the feature points matched with each auxiliary image, and the connection point error equation is constructed by combining the affine transformation model.

[0042] 2. Registration of internal laser height data and external laser height data

[0043] Gao Fen 7 carries a full-wave laser height meter and a footprint camera, which is used to obtain high-precision laser control points, i.e. internal laser height data. The footprint image has the characteristics of low resolution (3.2m), and how to register the footprint image and the stereo image is the key to successfully extracting the laser points, as shown in Figure 3 The application adopts the method of finding the conversion relationship between the footprint image and the stereo image to register the footprint image and the stereo image.

[0044] Firstly, the SIFT algorithm is used to extract feature points from the footprint image, which are used as target feature points, and then the coarse positioning method consistent with the connection point extraction is used to calculate the image block containing the target feature points on each stereo image,

[0045] Then, the image block obtained by coarse positioning is down-sampled to be consistent with the resolution of the footprint image, and the least squares matching algorithm is used to obtain the fine coordinates of the image block after down-sampling by taking the feature points on the footprint image as a template.

[0046] For the screening of laser points of GF-7 itself, the part of ECP_FLAG of 1 and 2 in the three-level product is selected as the laser points with higher quality. The homography matrix from the footprint image to the stereo image can be calculated from the matching point pairs between the footprint image and the stereo image. The image side of the laser point on the stereo image can be obtained by inputting the image side of the laser point on the footprint image into the homography matrix, and then the internal laser point error equation is constructed in combination with the affine transformation model.

[0047] In actual building damage assessment, the problem of uneven distribution of laser points may exist in the area to be assessed. The application also matches the external laser height data with the stereo image, and the specific steps are as follows:

[0048] For the screening of external laser height data, the laser points with a slope less than 2 are selected as the laser points with higher quality. Then the image coordinates I1 of the external laser height data are projected onto a single stereo image, the homography matrix is constructed according to the extraction results of the same named points in the above method, the homography matrix is used for coarse matching of the laser height data (I1 is input into the homography matrix to obtain the rough image coordinates I2, I 3, I4 of the laser height data on other stereo images), and finally the fine matching of the laser height data is performed by using the least square template matching on all stereo images respectively, and the external laser point error equation is constructed by combining the matching point pairs with the affine transformation model.

[0049] Step two, based on the above connection point error equation, internal and external laser point error equation, the adjustment model of the rational function imaging model based on stereo image is constructed, the adjustment model is solved, the regional network adjustment is completed, and the compensation coefficient corresponding to each stereo image is obtained, so as to obtain the adjusted rational function imaging model corresponding to each stereo image.

[0050] There is a relative positioning error between the stereo image pairs / groups themselves, and the error model is often constructed based on the rational function imaging model and the extracted same named point pairs / groups, and the error model is solved to finally eliminate the relative positioning error between the stereo images.

[0051] The rational function imaging model is a mathematical fitting form of the strict imaging model. In order to eliminate the errors caused by the satellite push-broom process, the inaccurate attitude and the orbit positioning, the research often uses the affine transformation model to compensate the rows and columns of the stereo satellite image to correct these errors. The adjustment model based on the rational function imaging model is constructed as follows:

[0052] V = AX1 + BX2 - L (1)

[0053] Wherein, V is error remainder; A is partial derivative of undetermined solving coefficient in affine transformation compensation model, X1 is undetermined coefficient compensation quantity corresponding to it; B is partial derivative of point object side, X2 is object side compensation quantity corresponding to it; L is actual image side residual difference after adding compensation model;

[0054] Suppose that the stereo images selected in the above text are respectively the rear view stereo image, the front view stereo image and the rear view stereo image, the front view stereo image before the building is shaken, and the same name points are respectively denoted as The compensation coefficients are respectively denoted as At this time, the error model is increased from the original double to quadruple, and the corresponding coefficient matrix A, B and residual matrix L in the error equation are changed to:

[0055]

[0056] Wherein, The partial derivative of the compensation form of the rational function imaging model to the affine compensation coefficient is represented as The partial derivative of the compensation form of the rational function imaging model to the object side of the same name point is represented as X RFM The image side is calculated by the rational function imaging model of the stereo image and the object side.

[0057] The above formula (1) is a general form of adjustment model, and in actual research, corresponding adjustment needs to be made for different points, such as the separation of plane and height control proposed by Tang Xinming. For laser point height measurement points, due to the high ranging accuracy but poor plane accuracy, the height value is often regarded as an accurate value, and the horizontal direction (longitude, latitude) is regarded as a value with error, that is, the partial derivative of the above adjustment model is not needed for the height, and only the partial derivative for the horizontal direction is needed.

[0058] The adjustment model of the application mainly includes the following three types:

[0059]

[0060] Among them, the first type of equation is the connection point error equation, the second type is the virtual control point error equation, and the third type is the laser point error equation. tp , V vcp , V las Respectively represent the error terms of the connection points, the virtual control points and the laser points, A tp , A vcp , A las Respectively represent the partial derivative terms of the undetermined solving coefficients in the affine transformation compensation model of the connection points, the virtual control points and the laser points, X aff Is the undetermined solving compensation coefficient, B tp , B vcp , Blas Represent the partial derivatives of the connection point, virtual control point and the object side of the laser point, X tp , X vcp , X las Represent the object items of connection points, virtual control points and laser points respectively, L tp , L vcp , L las Represent the residual terms of connection points, virtual control points and laser points respectively, P tp , P vcp , P las Represent the weight items of the connection point, virtual control point and laser point respectively. For different types of points, the unknown quantities to be solved in X are also different. For example, in the connection point error equation, the three unknown quantities of longitude, latitude and elevation need to be solved; but in the laser point error, due to the high vertical accuracy of the laser point but low horizontal accuracy, the longitude and latitude need to be solved, and the elevation is regarded as a fixed value. In addition, since the present invention does not require high-precision absolute horizontal control, only using the connection point and the laser point will lead to the ill-conditioning of the normal equation, so it is necessary to add a virtual control point to constrain the plane freedom. The average elevation of a single stereo image (obtained by the rational function positioning parameter) is used as the elevation of the virtual control point and the object space of the upper left corner of the stereo image and the ground resolution of the stereo image are used to obtain the longitude and latitude of the virtual control point to jointly determine the object space of the virtual control point. The stereo image is evenly divided into nine blocks, such as nine blocks, and a virtual control point is taken in each block. The error equation of the virtual control point is constructed using formula (1) to accelerate the convergence speed of the equation solution. For different types of points, the unknown quantities to be solved in X are also different. For example, in the tie point error equation, all three unknown quantities, longitude, latitude, and elevation, need to be solved. However, in the laser point error equation, due to the high vertical accuracy but low horizontal accuracy of laser points, both longitude and latitude need to be solved, and the elevation is considered a fixed value. In the virtual control point error equation, the object space is considered a fixed value. Since it is a hypothetical planar constraint, its weight is set to 1 / 100 of the tie point weight.

[0061] Step 3: Use the coarse-to-fine positioning method to automatically measure the corner points of the building and obtain the image coordinates corresponding to each corner point.

[0062] Buildings, such as houses, can show significant differences before and after an earthquake, such as deformation, dislodged tiles, and collapse. Furthermore, due to differences in the timing of the data acquisition, radiometric differences can be significant. By selecting the image space of the building where the changes occurred and applying a positioning model that eliminates relative positioning errors between stereo images, we can determine the building's object space. This object space is then back-projected onto another pair of stereo images. Visual interpretation can then yield sub-pixel-accurate image space for the building.

[0063] The imaging range of stereo images is large, and it is generally necessary to select fine building images through visual interpretation. This method requires comparison on multiple images, which is time-consuming and labor-intensive. Figure 4 As shown, the present invention proposes a coarse-to-fine automatic positioning method for fine measurement of house images.

[0064] First, the fine image-space coordinates of the corner points of the house roof are selected on the main image through visual interpretation. Combined with the homography matrix constructed by the connection points extracted above, that is, the kernel constraint, the coarse positioning of the house corner points on the other secondary images can be achieved. Then, the fine image-space coordinates of the house corner points on the main image are used as a template. Combined with the coarsely positioned image blocks, the least squares template matching algorithm is input and the points with correlation coefficients greater than 0.8 are retained. The fine image-space coordinates of the house corner points on the other secondary images can be automatically calculated.

[0065] It is worth pointing out that if the matching before and after the earthquake is successful, it means that the current house was not damaged before and after the earthquake; if the matching before and after the earthquake is unsuccessful, it means that the house has changed significantly before and after the earthquake and cannot be matched successfully. At this time, the homography matrix constructed by the connection points, that is, the transformation result of the kernel line constraint, can be directly used as the position of the collapsed house, that is, the image coordinates of the house including latitude and longitude information can be obtained.

[0066] Step 4: Input the image coordinates in step 3 into the rational function imaging model after adjustment of the stereo image to calculate the height change of the building and realize the assessment of the degree of building collapse.

[0067] The European Macroseismic Scale 1998 (EMS98) categorizes damage to buildings into five levels: slightly damaged, moderately damaged, severely damaged, very severely damaged, and collapsed. This study, incorporating the EMS98 and leveraging the maximum impact of remote sensing imagery, combined levels 2 through 4 into a single scale: slightly damaged, moderately damaged, and collapsed, as shown in the table below.

[0068]

[0069] (Excerpt from EMS98)

[0070] The height calculated using stereo images is based on WGS84, which is not the actual height of the house. Based on this, this study combined the height change ratio of the houses in the disaster area to calculate the damage degree of the houses. The damage degree assessment formula (6) is used. j Indicates the damage degree of the four corner points of a single house, where H pre ,H post are the measured heights of the houses before and after the earthquake, H groundFor the ground surface height near the house, formula (7) represents a specific calculation method of the number of collapsed house floors, wherein C represents the standard height of the local house building.

[0071] In order to fully verify the method of detecting three-dimensional changes of houses in disaster areas and evaluating house collapse by combining laser height measurement data with stereo images before and after the earthquake, the experimental selected stereo images of Gao Fen 7 taken in Oaxaca, Mexico. On June 23, 2020, a 7.4-magnitude earthquake occurred in this area, causing some house damage and casualties. Based on the two groups of Gao Fen 7 stereo images (4 stereo images) taken before and after the disaster, the method of processing the above-mentioned combined laser height measurement data and stereo images before and after the earthquake and the method of evaluating the damage degree of the house are verified, and the specific results are shown in Figure 5

[0072] In order to verify the method of joint adjustment before and after the earthquake, four image combinations are designed, which are: ① positioning alone before the earthquake ② positioning alone after the earthquake ③ using the pre-earthquake image alone after joint adjustment of the pre-earthquake and post-earthquake stereo images ④ using the post-earthquake image alone after joint adjustment of the pre-earthquake and post-earthquake stereo images. As shown in Table 1, in the case of controlled evaluation using laser height measurement data, compared with the case of positioning alone before the earthquake, after unifying the positioning accuracy before and after the earthquake, the height accuracy is obviously improved by 52%, and the vertical accuracy RMSE is improved from 2m to 0.98m; at the same time, in the case of no control, the positioning accuracy before the earthquake is improved by 82% after unifying the positioning accuracy before and after the earthquake, and the vertical accuracy RMSE is improved from 10.28m to 2.3m.

[0073] Table 1

[0074]

[0075]

[0076] The present application combines the disaster area report and the visual interpretation of the house damage within the stereo image range, selects four changed house areas A, B, C and D, which are distributed as shown in Figure 6 , and evaluates the damage degree of the house by using the method proposed in the foregoing, and the evaluation results are shown in Table 2, wherein the accuracy of area A reaches 93.6%, the accuracy of area B reaches 47%, the accuracy of area C reaches 97.3%, and the accuracy of area D reaches 100%, which meets the accuracy requirements of the evaluation of the house damage in the disaster area.

[0077] ​Table 2 gives the results of damaged / changed houses. Table 2 gives the number of detected house changes, the height of the house corner points and the changes thereof obtained by pre-seismic and post-seismic stereoscopic image positioning after joint adjustment of the pre-seismic and post-seismic stereoscopic images, the actual height, the actual number of floors, the estimated number of damaged floors and the estimated damage degree (intact, partially collapsed, completely collapsed, newly built).

[0078] In Table 2, the four corner points at A and B have large height changes before and after the earthquake, and the reduced height difference is almost the same, so they are detected as completely collapsed; the four corner points at C and D have obvious height increments after the earthquake compared with before the earthquake, which is used to assist the detection of collapsed houses, and they are also detected as newly built houses, which shows that the method proposed in the present application can detect the obvious height changes of collapsed or newly built houses.

[0079] Table 2

[0080]

[0081] The above-described and above-embodied examples are only used to illustrate the technical solutions of the present application, but not to limit the same; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can still be modified, or some technical features thereof can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A building collapse assessment method based on stereo images and laser altimetry data, characterized in that The following steps are involved: Step 1: Use a coarse-to-fine matching method to match multiple stereo images before and after the earthquake and build a connection point error equation in combination with an affine model. At the same time, internal laser altimetry data, i.e., footprint images, are matched with stereo images to extract laser points and build an internal laser point error equation in combination with an affine model. External laser altimetry data are matched with stereo images and build an external laser point error equation in combination with an affine model, and an adjustment model based on a rational function imaging model is jointly constructed. Step 2: Solve the adjustment model, complete the regional block adjustment and obtain the compensation coefficient corresponding to each stereo image, thereby obtaining the rational function imaging model after adjustment corresponding to each stereo image; Step 3: Use a coarse-to-fine positioning method to automatically measure the corner points of the building and obtain the image coordinates corresponding to each corner point; Step 4: Input the image coordinates obtained in step 3 into the rational function imaging model after adjustment of the stereo images before and after the earthquake, calculate the height change of the building, and realize the assessment of the degree of building collapse; The method for constructing the connection point error equation in step 1 includes the following steps: Step I: Select one of the multiple stereo images as the main image and the rest as secondary images, and use the SIFT algorithm to extract feature points from the main image; Step II: Calculate the longitude and latitude corresponding to each feature point on the main image, use this as the target feature point, combine the maximum and minimum elevation values ​​of the main image, input the rational function imaging model of the other secondary images, obtain the image block containing the target feature point, and then use the target feature point on the main image as a template and use the least squares matching algorithm to calculate the fine coordinates of the target feature point on the other secondary images; Step III: Use the target feature points of the primary image and the corresponding matching points of the other secondary images as connection points, and construct the connection point error equation in combination with the affine model; The method for constructing the internal laser point error equation in step 1 includes the following steps: Step i: Use SIFT to extract feature points from the footprint image; Step ii: Calculate the longitude and latitude corresponding to each feature point on the footprint image, use this as the target feature point, combine the maximum and minimum elevation values ​​of the footprint image, input the rational function imaging model of each stereo image, obtain the image block containing the target feature point, and then use the target feature point of the footprint image as a template and use the least squares matching algorithm to calculate the fine coordinates of the target feature point on each stereo image; Step III: Calculate the homography matrix from the footprint image to each stereo image based on the matching point pairs between the footprint image and each stereo image, input the image coordinates of the laser point on the footprint image into the matrix, obtain the image coordinates of the laser point on the stereo image, and construct the internal laser point error equation based on the affine model; The adjustment model based on the rational function imaging model is constructed in the step 1 as follows: Among them, the first type of equation is the connection point error equation, the second type is the virtual control point error equation, and the third type is the laser point error equation composed of internal and external. tp , V vcp , V las Represent the error terms of connection points, virtual control points and laser points respectively, A tp , A vcp , A las Represent the partial derivatives of the connection points, virtual control points and laser points to the undetermined solution coefficients in the affine transformation compensation model, X aff is the compensation coefficient to be determined, B tp , B vcp , B las They represent the partial derivatives of the connection point, virtual control point and the object side of the laser point, X tp , X vcp , X las Represent the object items of connection points, virtual control points and laser points respectively, L tp , L vcp , L las Represent the residual terms of connection points, virtual control points and laser points respectively, P tp , P vcp , P las Represent the weight items of connection points, virtual control points and laser points respectively.

2. The building collapse assessment method based on stereoscopic images and laser altimetry data according to claim 1, characterized in that: When the SIFT algorithm is used to extract feature points from the main image in step I, the main image is first divided into blocks, and then the SIFT algorithm is used to extract feature points from each image block. The ANMS algorithm is then used to homogenize the feature points in each image block, and feature points with response values ​​greater than 20% are retained.

3. The building collapse assessment method based on stereoscopic images and laser altimetry data according to claim 1, characterized in that: The external laser altimetry data is projected onto each stereo image, and then the corresponding homography matrix is ​​constructed using the connection points obtained according to claim 1 to coarsely align the laser altimetry data. Then, the laser altimetry data is finely aligned for all stereo images using least squares template matching. The external laser altimetry data and the matching points corresponding to each stereo image are combined with an affine model to construct an external laser point error equation.

4. The building collapse assessment method based on stereoscopic images and laser height measurement data according to claim 1, characterized in that: The automatic measurement of building corners is achieved in step three as follows: first, the fine image-space coordinates of the building corners are selected on the primary image by visual interpretation, and the corresponding homography matrix is ​​constructed in combination with the connection points obtained according to claim 1 to achieve coarse positioning of the building corners on other secondary images. Then, the fine image-space coordinates of the building corners on the primary image are used as a template, and a least squares matching algorithm is adopted to retain points with correlation coefficients greater than 0.8, and the fine image-space coordinates of the building corners on other secondary images are automatically calculated.

Images